We know how to solve the problem of software reliability. We have a rich body of formal methods that allow us to build perfectly reliable computer systems, up to the failure envelope of the hardware itself (from thermal, radiative, mechanical, etc. stress).
The reason we only use these techniques on certain critical systems is that the cost is somewhat higher than normal "unsafe" development and the incentives are not there yet.
Instead of turning our industry into a horrific morass of red tape, with questionable upside and unquestionable downside, let's just incentivize the use of correctness technology where appropriate.